The Impact of Improved Highways on Indian Firms

Saugato Datta�

World Bank Group

February 13, 2008

Abstract

India�s Golden Quadrilateral Program aimed at improving the quality and

width of existing highways connecting the four largest cities in India. This af-

fected the quality of highways available to �rms in cities that lie along the routes

of the four upgraded highways, while leaving the quality of highways available to

�rms in other cities una¤ected. This feature allows for a di¤erence-in-di¤erence

estimation strategy, implemented using data from the 2002 and 2005 rounds of

the World Bank Enterprise Surveys for India. Firms in cities a¤ected by the

Golden Quadrilateral highway project reported decreased transportation obsta-

cles to production, reduced average stock of input inventories (by about a week�s

worth of production), and a higher probability of having switched the supplier

who provided them with their primary input. Firms in cities where road quality

did not improve displayed no signi�cant changes.

�Email: sdatta1@worldbank.org. Address for Correspondence: 4K-244, 2121 Pennsylvania Avenue, Washington DC20036

1 Introduction and Motivation

This paper studies the response of �rms in India to an ambitious program of highway

improvements in order to measure the economic e¤ects of infrastructure investments, an

issue that has received widespread attention both in the context of developing countries

such as India and more generally. It is motivated by, and contributes to, two strands in

the literature.

The �rst is the literature on the role of the private sector in economic development,

and particularly e¤ects of government policies in enabling, constraining or retarding how

the private sector performs (for example, Djankov et al. 2008, Knittel 2002). Within this

literature, the Indian case has been the subject of a growing literature. Recent contri-

butions to this literature have argued, for example, that excessively restrictive licensing

policies were responsible for sub-optimal �rm performance (see Chari 2007). This paper

emphasizes the government�s role as an enabler for private-sector development, since it

�nds positive e¢ ciency e¤ects of large-scale investments in infrastructure. In the speci�c

case of India, this is a particularly compelling story, since many commentators have argued

that the �ipside of the Indian government�s historical over-involvement with regulating

and controlling private-sector activity was a tendency to neglect precisely those areas,

such as infrastructure provision, where it may have had the greatest scope for impact.

More generally, this paper examines the economic e¤ects of investments in infrastruc-

ture, in the spirit of papers such as Chandra and Thompson (2000). Highways are the

quintessential example of such investments, are often posited as being essential for higher

economic growth. Signi�cant budgetary resources are allocated to highway construction

in many countries based in part on a hypothesized causal link between infrastructure

development and economic growth. Yet, empirical evidence on this issue is �extremely

controversial, and consists of studies that are divided on both the magnitude and di-

rection of the net e¤ect of infrastructure spending on economic growth�(Chandra and

Thompson 2000).

Several issues complicate the estimation of the economic e¤ects of infrastructure. First,

while a variety of studies have found positive correlations between the level of infrastruc-

ture in an area and economic outcomes of interest, such as growth, endogeneous placement

of new infrastructure makes it di¢ cult to clearly quantify the causal e¤ects on economic

outcomes of interest. This issue can be summed up as follows: do areas with (more,

1

better) infrastructure show better economic outcomes because of the infrastructure, or do

better economic outcomes, or the potential for better economic outcomes, attract better

infrastructure? Thus, we may see that better roads are linked to faster growth, but if

we suspect that roads are likely to have been built or improved in areas that had the

highest potential for economic growth, our estimates of the impact of roads on growth

are biased upwards. Secondly, infrastructure can a¤ect not only the quantum, but also

the spatial distribution of economic activity, so that accounting for spillovers is impor-

tant if we try to capture the overall net impact of infrastructure. Thirdly, there is little

empirical evidence, apart from a small number of case studies (see, for example, Gulyani

2001), on the channels through which infrastructure a¤ects economic outcomes. While

studies of e¤ects on growth or employment exist, there is little information on �rm-level

variables (such as inventories, input costs, capacity utilization, competition) that may

respond to improvements in transport infrastructure, and through which the bene�cial

economic e¤ects of highway construction are channeled.

While truly random placement of a major infrastructure project is hard to conceive of,

a series of recent papers cognizant of the endogeneity issue have attempted to deal with

it in di¤erent ways. For example, Du�o and Pande (2004) use gradient to instrument

for the placement of hydroelectric projects in India, so as to obtain plausibly causal

estimates of the economic e¤ects of large hydroelectric projects. In the case of highway

construction, both Chandra and Thompson (2000) and Michaels (2007) use a feature

of the US Interstate Highway construction program that allows them to treat it as a

natural experiment for counties through which the new interstate highways (did not)

pass. The idea derives from the nature of the highway-building exercise: when a highway

is built to connect cities A and B, it must pass through areas that lie in between the two,

thus contributing to improved infrastructure in places that happen to lie in between the

(possibly endogenously chosen) points that the highway is built to connect. If there is

reason to believe that the precise route of the highway was not manipulated to include

some intermediate areas (counties, districts, cities) and exclude others based on factors

correlated with the outcomes of interest, then the highway construction can be treated as

exogenous to the areas that the highway runs through.

In this paper, I use an identi�cation strategy similar in spirit to that in Chandra

and Thompson (2000) to address the endogeneity issue. I argue that India�s Golden

Quadrilateral Program, which seeks to improve the quality and width of existing highways

2

connecting the four largest cities in India (Delhi, Mumbai, Kolkata, and Chennai), a¤ected

the quality of highway transportation available to �rms in cities that happened to lie

along the routes of the four National Highways it upgraded, while leaving the quality of

highways available to �rms in cities that lay along other national highways una¤ected.

This allows for a di¤erence-in-di¤erence estimation strategy, where changes in relevant

outcomes for a¤ected �rms are compared to the corresponding outcomes for �rms whose

location precluded their directly bene�ting from the highway program.

The structure and content of the data used here, which are from two rounds of the

World Bank�s Enterprise Surveys for India, are particularly well-suited for tackling the

questions discussed above. The panel structure of the data allows me to compare the

changes in the responses of a representative sample of 1,091 �rms in 37 cities - of which

19 were directly a¤ected by the highway upgradation program and the remaining 18 un-

a¤ected by it �between the year 2002, when the project had just begun, to the responses

of the same �rms in the year 2005, when it was approximately two-thirds complete. Fur-

ther, the Enterprise Survey data contain �rms�responses to questions which allow me to

directly measure how the choices that �rms make about inventories and input suppliers �

which the theory suggests should respond to better transport infrastructure (see for ex-

ample Shirley and Winston 2001) �are a¤ected by the quality of highway infrastructure.

The surveys also provide direct evidence about �rms�perceptions of the degree to which

transportation constitutes an obstacle to their productive activities. While these data are

di¢ cult to interpret in a cross-sectional setting, I utilize the panel structure of the data

to see how �rm�s responses to this question re�ected the change in road quality due to

the Golden Quadrilateral project.

The key �ndings of the paper are as follows. First, I �nd that �rms in cities a¤ected

by the highway project became 7.6 percentage points, or about 60 per cent, less likely to

report that transportation constituted a �major�or �severe�obstacle to production, while

there was no signi�cant change in the responses of �rms in o¤-Golden Quadrilateral cities.

Secondly, the di¤erences-in-di¤erences estimate of the e¤ect of the highway construction

on the average stock of input inventories held by a �rm is signi�cantly negative and

large in magnitude. Firms on the Golden Quadrilateral highways reduced their average

input inventory (measured in terms of the number of days of production the inventory

was su¢ cient for) by 7 days more than corresponding �rms which lay on other highways.

Thus, inventory management became sigini�cantly leaner for treatment �rms relative to

control �rms after the improved highways were put into place. Finally, I �nd that �rms

3

in cities that gained better highway access were much more likely to have switched the

supplier who provided them with their primary input than �rms in cities where road

quality did not improve. These latter two e¤ects can, I argue, be interpreted as showing

that �rms that gain access to higher-quality highways are able to produce more e¢ ciently

(as captured by their inventory management becoming leaner) than in the absence of the

improved highways, and to exploit opportunities that may earlier have been unfeasible,

which we argue are unambiguously economically bene�cial to these �rms. I �nd that

the results are robust to the exclusion of �rms in the 4 nodal metropolitan cities (Delhi,

Mumbai, Kolkata and Chennai), whose status as �on-Golden Quadrilateral�cities was a

matter of design rather than fortuitousness. Viewed together, these results support the

idea that the highway construction project eased the extent to, and channels through,

which transportation infrastructure constrains �rms.

However, I also �nd that although there was no initial di¤erence in the fraction of

�rms who report that transportation was a major obstacle between on- and o¤-Golden

Quadrilateral cities, the same is not true of inventory holdings: �rms in Golden Quadri-

lateral cities held signi�cantly higher input inventories before the highway project than

�rms in o¤-Golden Quadrilateral cities. While not directly contradicting the identifying

assumption of common trends, this does suggest that the argument made for using non-

Golden Quadrilateral cities as controls for cities directly a¤ected by the project may be

problematic: perhaps the highways that were included in the highway project were at

that point the most congested or overburdened. I attempt to understand why inventory

behavior was so di¤erent among �rms in Golden Quadrilateral cities compared to unaf-

fected �rms by including controls for industrial composition and also by revisiting the

perceptions data, but the results are somewhat contradictory. If anything, fewer �rms in

Golden Quadrilateral cities said that transportation was no obstacle or a minor obstacle

than �rms in non-Golden Quadrilateral cities before the highways were in place, which

does not square well with their reported inventory holdings.

The plan of the rest of the paper is as follows. Section 2 provides background infor-

mation on the speci�c context of this paper, the Indian highway network and the Golden

Quadrilateral project. Section 3 explains the identi�cation strategy used in the paper,

which relies on using the Golden Quadrilateral Project as a shock to the quality of high-

way infrastructure to cities it passed through versus those it bypassed that is unrelated

to (�rms in) the cities�(potential) economic performance. Section 4 motivates the focus

on inventory behavior and supplier relationships, and discusses the data and variables

4

used in this paper. Section 5 presents the key results and their interpretation. Section 6

concludes.

2 Background

2.1 The Indian Highway System

India�s road network, which carries 65 per cent of the country�s freight (World Bank 2007)

is dominated by its 66,000 kilometer-long network of 228 federally-built and maintained

National Highways, which connect major cities in the country to one another, and which

carry 40 per cent of the country�s road tra¢ c inspite of comprising only 2 per cent of

its aggregate road length of 3.3 millon kilometers. The prominent role played by the

National Highway system has motivated policymakers to pay considerable attention to

their quality and carrying capacity, which is widely acknowledged to be inadequate to the

needs of a rapidly-expanding economy, both in terms of lane capacity (the bulk of the

National Highway network as of the year 2000 was of two lanes or less; 32 per cent was in

fact single-laned; and virtually none of the network was of 4 or more lanes) and surface

quality. In addition, even national highways in India are not access-controlled, so that

highways often pass through busy towns and villages, with adverse consequences not only

for tra¢ c speed but also safety .

This has led to widespread congestion: the World Bank estimated that �a quarter of

all India�s highways are congested, reducing truck and bus speeds to 30-40 kmph�. It

is therefore no surprise that policy-makers argue that India�s highway network acts as a

constraint on highe economic growth: for example, the World Bank country strategy for

India �identi�es highway bottlenecks as one of the major constraints to poverty reduction

and private sector-led growth�(World Bank 2005).

2.2 The Golden Quadrilateral Project

In December 2000, the Government of India approved Phase I of the National Highway De-

velopment Program (NHDP), the most ambitious program of highway improvement since

Independence in 1947. Phase I of the NHDP was dominated by the Golden Quadrilateral

Program, under which the four national highways most directly linking the four largest

metropolitan cities in the country: NH2 (Delhi-Calcutta), NH8 (Delhi-Mumbai), NH4

(Mumbai-Chennai), and NH5(Calcutta-Chennai) were to be upgraded to �international-

standard�access-controlled 4- or 6-lane dual-carriageway highways with grade separators,

5

access roads, etc. Actual implementation of the Golden Quadrilateral Project began in

2002, by the end of which year contracts for most sections had been awarded, and was

substantially, though not completely, accomplished by the target completion date of De-

cember 2004: by 2006, 70 per cent of the project had been completed1.

The Golden Quadrilateral marked a distinct shift in India�s road-building strategy.

Prior to the approval of the NHDP, India had concentrated its road-building budget on

expanding all-weather access rather than improving quality or capacity. Prior to the

NHDP, virtually all of the 600,000 km that was added to India�s road network in the

1990s consisted of �very low-standard roads to reach more of the rural areas which had

before been outside the reach of all-weather access. High standard arterial highways ...

were largely neglected�(World Bank 2005:2). The highways upgraded under the Golden

Quadrilateral project were therefore India�s �rst set of international-quality arterial high-

ways, and the Golden Quadrilateral project thus led to a sharp increase in highway quality

for the areas it connected over and above any other road improvements that may have

taken place (because of other government programs involving road resurfacing, new road

construction, etc.).

3 Identi�cation Stragegy

3.1 The Golden Quadrilateral Project as a Natural Experiment

As discussed above, the key issue that makes it di¢ cult to estimate the causal e¤ect of

infrastructure such as highways is that their placement is non-random, so that places that

receive more or better infrastructure are systematically di¤erent from areas that do not.

However, as Chandra and Thompson (2000) have pointed out, the nature of many large

highway network projects is such as to allow them to be used as an exogenous shock to

areas that they pass through. The argument relies on the fact that when highways are

built to connect two places, they pass through areas that lie between those two points.

If the highway route is not determined to speci�cally pass through certain intermediate

areas to the exclusion of other possible places it could have been routed through, then it

may be argued that intermediate areas get better infrastructure not as a consequence of

economic or other characteristics, but merely because of where they are located.1Phase 1 of the NHDP also included 931 kilometers of the proposed North-South-East-West (NSWEW) Highways, but

this part of the NHDP was eventually e¤ectively pushed into Phase II: as of 2007, only (xx) per cent of the length of theNSEW Highways had been completed.

6

The argument for using the Golden Quadrilateral project in a natural experiment

framework is similar to that in Chandra and Thompson (2000), who use the US In-

terstate Highway construction program as an exogenous shock to the connectedness of

districts through which the highways built passed relative to those they bypassed. The

Government�s decision to pick the existing highways connecting the country�s four largest

cities as the ones that were to be widened and improved as part of the project meant

di¤erent things for smaller cities in the country depending on which national highway a

particular city lay on. All major cities in India lie on the route of at least one National

Highway, so whether or not a city saw a sharp improvement in the quality of its highway

links depended on which pre-existing National Highway passed through or near it. Cities

were not included or excluded from bene�ting from the highway improvements under the

Golden Quadrilateral based on their existing or potential economic growth, as we might

imagine would be the case if the route of the highways was not pre-determined. The

highways included in the Golden Quadrilateral provided the most direct links between

the country�s four largest cities, and the highway routes were not realigned to exclude

some cities and include others.

Thus, cities could not opt in or out of getting improved highway access, thus precluding

a situation where the highway was routed through areas that �needed it most�(for example,

areas which were growing more rapidly than others, or which had particularly pressing

infrastructure issues). A city that lay between Delhi and Calcutta but was not on NH2

(such as Lucknow, a city in the state of Uttar Pradesh) would not directly bene�t from

improved connectivity from the Golden Quadrilateral, but a city that did lie on NH2 (such

as Kanpur, a similar-sized city in the state of Uttar Pradesh) would. This motivates the

empirical strategy used in this paper, where the changes in outcomes of �rms that lay

in cities that lay along the upgraded highways (�treated� cities) are compared to the

corresponding changes in outcomes for �rms in cities that did not lie along the route

of the Golden Quadrilateral highways (�control cities�) in order to measure the e¤ect of

highway quality. Put di¤erently, our estimate of the e¤ect of the Golden Quadrilateral

uses the changes in outcome variables for �rms on non-Golden Quadrilateral highways as

estimates of the counterfactual for treated �rms.

A caveat is, however, in order. The validity of the identi�cation strategy relies on the

assumption that the �rms on the Golden Quadrilateral highways and other �rms in the

sample have common trends in inventory and supplier behavior, which is what makes the

change over time in the outcomes for the latter sets of �rms a good control for the former

7

set. It is not possible, given that there is only one pre-treatment period, to test this

common trends assumption; however, I will explicitly control for pre-existing di¤erences

between �rms in treatment and control cities in the regressions whose results are pre-

sented here. As will be discussed in Section 5.4, a potential confound comes from the fact

that while cities could not self-select onto the routes of the highways that were upgraded,

the choice of which highways to upgrade may have been related to their existing condi-

tion, congestion conditions, tra¢ c requirements, etc. In that sense, it is possible that our

di¤erence-in-di¤erence estimates of the impact of the Golden Quadrilateral Project are

upward biased, because road conditions along what eventually became the Golden Quadri-

lateral were initially worse than road conditions for non-Golden Quadrilateral cities.

4 Data and Outcome Variables of Interest

4.1 Data Sources and Contents

The outcome data used in this paper are from two rounds of the World Bank Enterprise

Surveys carried out worldwide in the years 2002 and 2005. A subset of Indian �rms in

37 Indian cities were surveyed in both years, giving us a panel of �rms whose responses

can be compared across two years in order to measure how the improvements in road

infrastructure carried out as part of the Golden Quadrilateral a¤ected �rms who were

on the route of the Golden Quadrilateral relative to those that were not. Of the cities

covered in both survey years, 19 cities lay along the four highways that are part of the

Golden Quadrilateral: NH2 (connecting Delhi and Calcutta), NH8 (connecting Delhi and

Mumbai), NH4 (connecting Mumbai and Chennai) and NH5 (connecting Chennai and

Kolkata), while the remaining 18 lay on other National Highways. Data on the location

of the cities in the sample relative to Golden Quadrilateral and other highways is from

NHAI maps supplemented with satellite data from Google Maps, which allows me to

pinpoint which highway(s) a particular city lay on, and thus to classify the cities in the

sample into on-Golden Quadrilateral (treated) versus o¤-Golden Quadrilateral (control)

cities.

4.2 Outcomes of Interest: Inventories, Supplier Relationships and Percep-

tions of Transportation Quality

If infrastructure constrains �rms�choices, we should �nd evidence that removing an in-

frastructural constraint leads to behavior that is unambiguously �better�for �rms than

8

the choices they were making before the constraint was relaxed. Here, I focus on aspects

of �rms�inventory and supplier behavior, which I argue below are plausibly constrained

by inadequate highway infrastructure, and which would therefore reasonably be expected

to respond to the relaxation of the transport constraint; these are also aspects of �rm

behavior that the data allow me to examine. I also examine responses to a question in

the surveys that asked �rms to rate how much of an obstacle transportation constituted

for them as an added source of information, albeit perceptions-based, on how �rms�per-

ceptions of the adequacy of the available tranportation infrastructure responded to the

better highways.

4.2.1 Inventories

The intuition behind a focus on input inventories is straightforward. Since �rms hold

input inventories in order to ensure a smooth production �ow, a �rm�s target inventory

level is increasing in its expectations of future sales. However, �rms incur costs (capital,

storage and depreciation costs, among others) from holding inventories, also increasing in

the amount of inventory held. The balance between these considerations a¤ect the �rm�s

reorder point, the minimum sustainable level of inventory.

As Shirley and Winston (2001) point out, faster and more reliable highway transporta-

tion means that input orders will be received more quickly and with less uncertainty. This

means that, conditional on its expectations about production and sales, the �rm will need

to keep lower input inventories at any given point in time. The intuition is clari�ed by

thinking of the extreme case, where orders could be received instantaneously, in which

case �rms would not have to keep any inventory at all: �rms could practice what is known

as �Just-in-Time�inventory management. Ceteris paribus, therefore, we should expect a

negative relationship between the level of inventories and the availability and quality of

transportation, of which highway connectivity is a component.

The data used here allow both this hypothesis to be tested, because both the 2002

and 2005 surveys asked the respondent about inventory stocks held by the �rm. Firms

were asked how many days of production�s worth of their most important input they

held inventories of at the time a new delivery of this input came in from their suppliers.

The answer to this question is thus our measure of �rms�inventory holdings. Using the

location of the �rm in order to classify it as a �treatment� or �control��rm, I seek to

measure whether on-Golden Quadrilateral �rms saw greater reductions in their reported

9

input inventory holdings between 2002 and 2005 relative to o¤-Golden Quadrilateral �rms.

4.2.2 Supplier Relationships

Another plausible way in which improved roads could impact �rms�input use behavior

is by increasing the feasible set of suppliers from whom they could purchase the inputs

needed for production. A supplier who was earlier not a feasible choice due to a combina-

tion of distance and poor road quality could become feasible given better roads. However,

there would be no such expansion in the choice set of �rms who were una¤ected by im-

provements in road quality. If it were the case that road quality was constraining �rms

from making their optimal supplier choices, then �rms in cities impacted by the highway-

upgrading program should be more likely to re-optimize their choice of input supplier than

�rms in control cities, whose choice set is una¤ected. On the other hand, any supplier

who was �feasible�before the highway upgrade continues to be feasible after the highway

project, so that if a �rm does not change its supplier, this implies that the pre-existing

choice continues to dominate its expanded choice set, so that road quality was not a

binding constraint.

Firms were asked about the length of time they had been in business with their main

input supplier, so that evidence of di¤erential changes in supplier turnover come from

di¤erential changes in the length of the supplier relationship. Consider how the change

in the length of the relationship reported in the two surveys would relate to supplier

turnover. The greater the propensity of �rms in any sample of �rms to change suppliers,

the lower would be the mean increase in the length of the supplier relationship between

the two time periods. Thus, the magnitude of the increase in the supplier relationship

length allows us to make inferences about the probability of supplier change between the

two periods. Extending this idea to two comparable groups of �rms, therefore, we would

�nd that if more �rms in Group A had changed suppliers than in Group B, then the mean

di¤erence in the length of the relationship between the survey years would be smaller in

Group A than in Group B. The di¤erences-in-di¤erences estimate of the e¤ect of the

Golden Quadrilateral project on supplier relationship length then allows us to isolate the

e¤ect of the project on supplier turnover, with a negative coe¢ cient implying that �rms

in the treatment group increased their supplier relationship length by less than �rms in

the control group, implying that treated �rms were more likely to have acquired a new

supplier in the period between the surveys than control �rms.

10

4.2.3 Perceptions of Transportation as an Obstacle to Production

Finally, the data also provide us with information about surveyed �rms�perceptions about

how severe an obstacle transportation constituted for them. Firms were asked to choose

whether transport constituted no obstacle, a minor obstacle, a moderate obstacle, a major

obstacle, or a very severe obtacle to production. In the results presented here, I reclassify

this �ve-point scale into three categories of how severe an obstacle transportation con-

stituted to the �rm in question: low (incorporating no obstacle or a minor obstacle, the

bottom two categories); medium (which corresponds to the �moderate obstacle�answer),

and high (which incorporates the highest two degrees in the data, major and very severe

obstacle). There are well known problems with perceptions data, which make it di¢ cult

to compare the responses of di¤erent �rms to each other, but this is obviated to some

degree by being able to look at changes in the same set of �rms�perceptions over time,

which is what the panel structure of the data allows me to do. This is the approach

adopted here in order to provide a set of results which supplement those from the data

on inventory stocks and supplier turnover.

5 Results

Table 1 presents the industrial composition of �rms in the sample, broken up by the

status (on or o¤ the Golden Quadrilateral) of the city the �rms were located in. The

distribution of �rms across industries in treatment and control �rms is similar: of the 8

top industries in the sample for treatment cities, 7 are also the among the top 8 in control

cities. Nonetheless, there are di¤erences: auto components and drugs/pharmaceuticals

are a greater proportion of �rms in treatment cities, while electronics �rms make up a

larger share of the control sample. In some of the regression results that follow, therefore,

I will control for industrial composition.

5.1 Input Inventories: Leaner Inventory Management

Table 2 presents the means and standard deviations of the inventory stock variable for on-

and o¤-Golden Quadrilateral cities in 2002 (the pre-period) and 2005 (the post-period).

A comparison between the 2002 and 2005 inventory holding �gures reveal there is a

large change in the amount of inventory held by �rms in cities a¤ected by the Golden

Quadrilateral project between the two survey years, while there is no such change in the

inventory holdings of �rms in the other cities in the sample. Firms in cities that lay on

11

the Golden Quadrilateral held approximately 7 days of production�s worth less of input

inventories in 2005 than they did in 2002, whereas �rms o¤ the Golden Quadrilateral

report virtually no change.

Table 3 presents results from regressions with �rms�inventory holdings as the depen-

dent variable. The coe¢ cient of interest is that on the post-treatment dummy (i.e. the

interaction term between the post-period and being in an on-project city), which is the

di¤erences-in-di¤erences estimate of the impact of the Golden Quadrilateral. The �rst

column presents results from the baseline speci�cation with no additional controls. The

estimated coe¢ cient on the interaction of the post-period dummy with the treatment

dummy is signi�cantly negative, implying that treatment �rms saw a reduction in their

reported inventory holdings of approximately 7.5 days worth of production over and above

that seen in the case of control �rms.

A possible complication is that the distribution of �rms across industries may have

di¤ered between treatment and control cities. If some industries then saw improvements

in inventory management technology or a reduction in input inventory requirements that

were unrelated to transportation infrastructure, and if these industries were dispropor-

tionately represented in cities on the Golden Quadrilateral, then the interpretation above

could be misleading. To check for this, Column II of Table 3 reports results from a spec-

i�cation with industry �xed e¤ects, which does not change the results: the coe¢ cient of

interest is -7.4, and is signi�cantly negative. Finally, Column III reports results from a

speci�cation with �rm �xed e¤ects. The coe¢ cient of interest continues to be signi�cantly

negative, and its magnitude is only slightly smaller, at -6.

5.2 Di¤erential Supplier Turnover: Evidence on the Length of Supplier Re-

lationships

As Table 4 shows, �rms in cities on the Golden Quadrilateral showed a signi�cantly lower

increase in the length of their relationship with their main input supplier, with the

point estimate of the coe¢ cient of interest varying between -0.21 and -0.24 in alternative

speci�cations. As I argued in Section 4.2.3 above, if supplier turnover were similar across

the two groups of �rms, then the interaction of the post-period with the treatment dummy

should not pick up a signi�cant coe¢ cient. The coe¢ cient on this variable thus suggests

that supplier turnover was greater for �rms on the Golden Quadrilateral than for other

�rms in the sample: �rms in cities on the Golden Quadrilateral increased the length

12

of their relationship with their main input supplier by around 4 months less than did

corresponding �rms in non-Golden Quadrilateral cities, thus implying that more �rms in

Golden Quadrilateral cities switched suppliers between the survey years than did �rms in

control cities.

5.3 Transportation as an Obstacle to Production

The fraction of �rms who said that transport constituted a �major�or a very severe ob-

stacle�to production, i.e. those that picked either the highest or second-highest ordinal

value on a 5-point scale changed di¤erentially between �rms in the a¤ected and una¤ected

cities, again con�rming the picture of disproportionate reductions in transportation ob-

stacles to �rms in the former set of cities. This percentage went down by 66.3 per cent,

from 12.09 per cent in 2002 to 4.27 per cent in 2005, but there was no corresponding

change for �rms in o¤-Golden Quadrilateral cities, where the 2002 and 2005 numbers

were 10.27 and 11.09 per cent of �rms, respectively. Table 5 presents the results of

di¤erences-in-di¤erences estimates: the results con�rm the results from the regressions

using inventories as the dependent variable, with the coe¢ cient of interest (that on the

variable post-treatment) being signi�cantly negative. As Columns II and III of Table 5

show, these results are robust to the inclusion of industry and �rm �xed e¤ects. The per-

ceptions data thus back up the results from the �gures on input inventory management

and the length of the supplier relationship presented in the preceding sub-sections.

All these results are as expected in that the Golden Quadrilateral eased transporta-

tion constraints di¤erentially for a¤ected �rms relative to other �rms. But were these

constraints equally severe for both sets of �rms to begin with, as one would expect if as-

signment was, as argued, unrelated to the conditions of the cities in either subset? Here,

the perceptions data and the inventories data no longer tell the same story.

5.4 Pre-Existing Di¤erences between Cities On and O¤ the Golden Quadri-

lateral

The coe¢ cients on the variable �treatment�, which is a dummy which takes the value 1 for

cities on the Golden Quadrilateral and 0 otherwise, is signi�cantly positive (see Table 3),

suggesting that there were signi�cant pre-existing di¤erences in inventory holding patterns

between �rms in treatment and control cities. A comparison across the rows of Column I of

Table 2 con�rms that non-Golden Quadrilateral cities held many more days of inventory in

13

2002 than Golden Quadrilateral cities did, suggesting either that road conditions or �rms�

characteristics/needs for inventories, or both, were systematically di¤erent between the

two kinds of cities prior to the project being initiated. In the absence of more pre-Golden

Quadrilateral data, it is not possible to explicitly test the �common trends�assumption

underlying the di¤erences-in-di¤erences estimation strategy, which may still be valid, but

the �nding reported above is potentially problematic for the simple exogenous change in

road quality story stressed here, as it implies that while it may be the case that individual

cities did not have the option to opt in or out of the project, there may be some reason

that the 4 highways included in the project were included: perhaps these were the most

congested and overburdened highways in the country. This is potentially problematic for

the interpretation of the magnitude of the di¤erences-in-di¤erences results presented thus

far, since it suggests that using the change in the inventory holdings of �rms in non-Golden

Quadrilateral cities may not be a good control for the counterfactual change in inventory

holdings in Golden Quadrilateral cities.

Is it then the case that while treatment cities saw an improvement in their transport

conditions that control cities did not, but began from a much worse position? Some

evidence that speaks to this question is available from �rms�responses in 2002 (the pre-

period) to the question about the degree to which transportation posed an obstacle to their

production. The distribution of �rms�responses are very similar in treatment and control

cities: about three-quarters of �rms in either set of cities say that transportation posed

either no or a minor obstacle to their production; likewise, 12 per cent of �rms in treatment

cities and 10 per cent in control cities say that it was a major or very severe obstacle.

If anything, as Table 6 shows, the fraction of �rms in treatment cities who say that

transportation poses no obstacle is larger, at 0.47, than the corresponding fraction (0.40) in

control cities. This does not suggest that �rms in treatment cities were signi�cantly more

disadvantaged when it came to transportation than �rms in control cities; if anything,

quite the opposite. However, it is worth keeping in mind that this relies on comparing

perceptions data across di¤erent �rms, an undertaking that is problematic to the extent

that two �rms may perceive the same quality of infrastructure very di¤erently: indeed, it

appears here that �rms in cities which would later bene�t from better highways rated the

quality of their transportation infrastructure very similarly to other �rms although they

also seem to have been holding far many more days�production worth of input inventories,

which suggests that perceptions may be quite clearly conditioned by what is expected or

appears to be the norm. The fact remains, however, that cities on and o¤ the upgraded

14

highways reported holding very di¤erent amounts of inventories prior to the project, so

that the magnitude of estimates of the project�s impact should be interpreted with some

caution; this is something that data limitations preclude a deeper investigation of at this

point.

5.5 Robustness Check: Excluding Nodal Metropolitan Cities

The results presented so far have all used data for all cities in the sample. Four of these

cities - Delhi, Mumbai, Chennai and Kolkata - are the metropolitan cities which form

the nodes of the improved highway system, and which the highways that were upgraded

were chosen to connect. One obvious worry therefore is that these cities did not receive

upgraded highways by virtue of lying on highways that were chosen to be upgraded:

that argument works, strictly speaking, only for other cities that lay on the upgraded

highways. I therefore test for the robustness of my results to the exclusion of these

metropolitan cities by running the regressions again with these cities excluded from the

sample. The results of the regressions on this non-metropolitan sample are presented in

Table 7 for the three key outcome variables used here - days of inventory held, supplier

history, and the probability that the �rm rated transport as a �major�or �very severe�

obstacle to production. Reassuringly, the exclusion of the nodal metropolitan cities has

no e¤ect on either the signi�cance, direction, and in particular the magnitudes, of the

estimated e¤ects of the highway program.

6 Conclusion

Firms in cities that lay along one of the four national highways connecting the four largest

cities in India that the Indian government upgraded as part of its Golden Quadrilateral

report holding about a week�s worth of production less of input inventories in 2005, when

much of the project had been implemented than in 2002, when work had just begun,

while �rms which lay in cities o¤ the Golden Quadrilateral highways report no such

change.These �rms also became much less likely to report that transportation was a

major or severe obstacle to obstacle to production in 2005 relative to their responses

to the same question in 2002. Firms on the upgraded highways also show a greater

propensity to change suppliers between the two years, suggesting that they found more

suitable suppliers at a greater rate than �rms in cities una¤ected by the highway project.

Seen together, these pieces of evidence substantiate the idea that improved highways

facilitated productive choices which �rms may have wanted to make even earlier, but

15

were constrained from being able to make by the quality of highways available to them.

There is a large literature (see for example Djankov et al. 2008, Knittel 2002) that

explores the e¤ect of policy variables, such as regulatory practices and taxation, on �rms�

productive choices and through these on �rm e¢ ciency. This paper contributes to this

literature by arguing that infrastructure quality is another key factor that a¤ects how

e¢ ciently �rms are able to produce, and by identifying the channels through which the

e¤ects of improved roads are channeled. In the speci�c context of India, the �ndings of

this paper, seen in conjunction with the �ndings of papers such as Chari (2007) contribute

to an understanding of what the government should (not) try to do if the private sector

in that country is to ful�l its potential as a driver of economic growth.

16

ReferencesChandra, Amitabh and E. Thompson. 2000. �Does Public Infrastructure A¤ect Eco-

nomic Activity? Evidence from the rural interstate highway system�. Regional Science

and Urban Economics 30(2000): 457-90

Chari, Amalavoyal. 2007. �License Reform in India: Theory and Evidence�. Mimeo,

Yale.

Demurger, Sylvie. 2001. �Infrastructure Development and Economic Growth: An

Explanation for Regional Disparities in China�. Journal of Comparative Economics. 29:

95-117.

Djankov, Simeon, Tim Ganser, Caralee Mcliesh, Rita Ramalho and Andre Shleifer.

2008. �The E¤ect of Corporate Taxes on Investment and Entrepreneurship�. NBER

Working Paper No. 13756. Cambridge, MA.

Du�o, Esther and Rohini Pande. �Dams�. MIT Department of Economics, Mimeo.

Fernald, J.G. 1998. �Roads to Prosperity? Assessing the Link between Public Capital

and Productivity�. AER 89:619-38.

Gulyani, Sumila. 2001. �E¤ects of Poor transportation on Lean Production and In-

dustrial Clustering: Evidence from the Indian Auto Industry�. World Development 29(7):

1157-77

Holl, Adelheid. 2004. �Manufacturing location and impacts of road transport in-

frastructure: empirical evidence from Spain�. Regional Science and Urban Economics.

34: 341-63

Knittel, Christopher. 2002. �Alternative Regulatory Methods and Firm E¢ ciency:

Stochastic Frontier Evidence from the U.S. Electricity Industry�. Review of Economics

and Statistics 84(3).

Michaels, Guy. 2007. �The E¤ect of Trade on the Demand for Skill �Evidence from

the Interstate Highway System�. Mimeo, LSE

Munnell, A.H. 1980. �Policy Watch: Infrastructure Investment and Economic Growth�.

Journal of Economic Perspectives 6(4): 189-98

Rephann, T.J. and A. Isserman. 1994. �New Highways as Economic Development

Tools: An evaluation using quasi-experimental matching

methods�Regional Science and Urban Economics 24, 723-51

Shirley, Chad and Cli¤ord Winston. 2004. �Firm Inventory Behavior and the returns

from highway infrastructure investments�. Journal of Urban Economics 55: 398-415

Winston, C., T. Corsi, C. Grimm, C .Evans. 1990. �The Economic E¤ects of Surface

17

Freight Deregulation�, Brookings Institution, Washington D.C

Wooldridge, Je¤rey. 2002. Econometric Analysis of Cross-Section and Panel Data.

Cambridge. MA: MIT Press.

World Bank. �Highway and Railway Development in India and China, 1992-2002�.

Transport Note, Energy and Infrastructure Unit, South Asia Region, The World Bank.

World Bank. 2007. India�s Transport Sector: The Challenges Ahead. World Bank,

Washington DC

18

Table 1: Distribution of Firms Across Treatment and Control Samples

Non-GQ GQ citiesNo. of Firms 487 604

Industrial Distribution (Per Cent of Firms in Sample in Each Industry)

Garments 12.94 9.77Textiles 12.94 12.58Drugs and pharma 7.39 13.41Auto components 9.86 14.57Electronics 3.7 7.12Electrical appliances 8.21 8.77Machine tools 6.57 6.62Food processing 12.94 5.13Leather 2.46 4.14Metals 8.01 3.48Plastics 5.13 4.3Rubber 2.05 1.66Paper 1.03 1.49Paints 0.62 0.5Cosmetics 0.41 0.17Other chemicals 4.11 3.97Marine food processing 0.82 0.33Agro processing 0.41 0.99Wood 0.41 0.17Sugar 0 0.66Mineral processing 0 0.17

Table 2: Days of Inventory Held

_____________________________________________________________2002 2005 Change, 2002-5

Non-GQ cities

Days of Inventory Held 22.87 22.82 -0.05(S.E.) (28.42) (19.31)

GQ cities

Days of Inventory Held 34.28 26.71 -7.57(S.E.) (47.66) (28.15)

_____________________________________________________________

Table 3: Effect of the Golden Quadrilateral on Days of Inventory Held

Dependent Variable: Number of Days of Inventory of Most Important Input

_____________________________________________

post_treatment -7.52*** -7.49*** -7.18*** -6.01***SE (2.79) 2.79 2.83 2.43t-ratio -2.69 -2.68 -2.54 -2.48

post -0.05 -.050 -0.12 0.094SE 1.59 1.58 2.10 1.51t-ratio -0.03 -0.03 -0.06 0.06

treatment 11.41*** 11.08*** -- --SE 2.39 2.38 -- --t-ratio 4.77 4.64 -- --

Industry dummies No Yes Yes YesCity fixed effects No No Yes NoFirm Fixed Effects No No No Yes

N 2109 2109 2109 2109______________________________________________

Table 4: Effect of the Golden Quadrilateral on Length of Firms' Relationship with Supplier

Dependent Variable: No. of Years In Business With Main Input Supplier

I II III IV___________________________________________________

post_treatment -0.238*** -.236*** -0.236*** -0.215***SE 0.089 .088 0.083 0.084t-ratio -2.68 -2.65 -2.85 -2.57

post 0.364 0.363 0.363 0.354SE 0.078 0.077 0.071 0.073t-ratio 4.660 4.700 5.120 4.880

treatment 0.478*** 0.461*** -- --SE 0.072 .0709709 -- --t-ratio 6.630 6.480 -- --

Industry dummies No Yes Yes NoCity fixed effects No No Yes NoFirm Fixed Effects No No No Yes

N 2182 2182 2182 2182___________________________________________________

Table 5: Effect of the Golden Quadrilateral on Firms' Perceptions of Transport

Dependent Variable: Dummy for Firm Saying Transport 'Severe' or 'Major' Obstacle

I II III IV___________________________________________________

post_treatment -0.084* -.084*** -0.084*** -0.084***SE 0.049 .025 0.025 0.024t-ratio -1.71 -3.34 -3.44 -3.48

post 0.008 0.008 0.008 0.008SE 0.043 0.020 0.018 0.019t-ratio 0.19 0.42 0.45 0.44

treatment 0.018 0.018 -- --SE 0.035 .0193 -- --t-ratio 0.52 0.94 -- --

Industry dummies No Yes Yes NoCity fixed effects No No Yes NoFirm Fixed Effects No No No Yes

N 2182 2182 2182 2182___________________________________________________

Table 6 : Percentage of Firms Reporting That Transport Was No Obstacle to Production

_____________________________________________________________2002 2005 Change, 2002-5

Non-GQ cities

% of Firms 40.22 60.29 20.07

GQ cities

% of Firms 47.23 67.89 20.66

_____________________________________________________________

Table 7: Effect of the Golden Quadrilateral Excluding 4 Nodal Cities

Dependent Variable: See Column Head

Sample: Firms in all cities except Delhi, Mumbai, Kolkata, and Chennai_____________________________________________

Inventory Suppliers Obstacle

post_treatment -7.791*** -0.196** -0.075***SE 2.728 0.090 0.027t-ratio -2.86 -2.19 -2.82

post 0.094 0.353*** 0.008SE 1.855 0.061 0.018t-ratio 0.1 5.8 0.5

Firm Fixed Effects Yes Yes Yes

N 1756 1814 1814______________________________________________